Data Similarity and Importance Using Local and Global Evidence Scores

1. A method for determining signal similarity between a query signal and a reference signal via composition, the method comprising: searching said reference signal to detect at least one matching region to a region of said query signal, the shape and size of said matching region being determined as a function of a local evidence score for said region, wherein said local evidence score is a function at least of a quality of the match of said region and of its statistical significance; and composing at least part of said query signal from at least one said matching region which provides at least one of: local evidence scores and a global evidence score for said query signal; wherein said global evidence score defines the extent of a global similarity of said query signal to said reference signal.

2. The method according to claim 1 and also comprising at least one of the following: determining irregular regions as regions whose said local evidence scores are below a threshold; and determining saliency within said query signal at least from said local evidence scores.

3. The method according to claim 1 and also comprising at least one of the following: employing at least said global evidence score for clustering a multiplicity of signals; clustering said regions within said query signal using at least said local evidence scores; and segmenting said query signal using at least said local evidence scores; and marking regions of said query signal with markings from said regions of said reference signal using at least said local evidence scores.

4. The method according to claim 1 and also comprising marking regions of said query signal with markings from said regions of said reference signal using at least said local evidence scores, where said markings comprise at least one of: labels, annotations, web-links, drawings, hyper-links, comments, graphical signs, XML data, markup language data, commercials advertisements, news, relevant tasks, appointments, events, references, citations, talkbacks, contact information, financial data, blogs, newsgroup links, RSS links, bookmarks services, email information, and highlights.

5. The method according to claim 1 and also comprising classifying items in at least portions of said query signal from at least said global evidence scores wherein said items are objects when said query signal is at least one of: an image and a video sequence, said items are actions when said query signal is a video sequence and said items are speech when said query signal is an audio signal.

6. The method according to claim 1 and also comprising recognizing items in said query signal as items in said reference signal using at least said global evidence scores, wherein said items are objects when said query signal is an image, said items are speech when said query signal is an audio signal and said items are speakers when said query signal is an audio signal.

7. The method according to claim 1 and wherein said signals are at least one of: a signal, a group of signals and a portion of a signal.

8. The method according to claim 1 and wherein said query signal is one portion of an input signal and said reference signal is another portion of said input signal.

9. The method according to claim 1 and wherein said query signal and said reference signal are at least one of the following types: a visual signal, an image, a video sequence, an acoustic signal, an electrical signal, a radar signal, a seismic signal, a audio signal, a sonar signal, a thermal signal, a medical signal, a medical image, functional magnetic resonance imaging (FMRI) data, an ultrasound signal, an electrocardiogram signal, x-ray data, range data, multi-spectral data, DNA sequence, amino acid sequence, protein data, text data, and an N-dimensional signal where N is an integer.

10. The method according to claim 1 and wherein said searching comprises: for each point q in said query signal, determining candidate small regions in said reference signal which are similar to a small region in the vicinity of point q; for each candidate region of q, maximally expanding each said candidate region while increasing said local evidence score of said region; and for each point q, choosing its region as the expanded region with the maximal said local evidence score.

11. The method according to claim 1 and wherein said regions are represented by an ensemble of patch descriptors.

12. The method according to claim 1 and wherein said local evidence score LES is defined as: LES ⁡ ( R | H ref ) = log 2 ⁡ ( LR ⁡ ( R ) ) LR ⁡ ( R ) = P ⁡ ( R | H ref ⁢ ) P ⁡ ( R | H 0 ) = ∑ T ⁢ P ⁡ ( R | T , H ref ) ⁢ P ⁡ ( T | H ref ) P ⁡ ( R | H 0 ) where P indicates a probability function, H ref is the hypothesis that a region R was generated by reference REF, H 0 is the null hypothesis and T is a transformation of a region of REF.

13. The method according to claim 12 and wherein said local evidence score is defined as a point evidence score PES, as follows: PES ⁡ ( q : H ref ) = max R , s . t . q ∈ R ⁢ LES ⁡ ( R | H ref )  R  where H ref is the hypothesis that a region R was generated by reference REF, and |R| is the size of R.

14. The method according to claim 1 and wherein said local evidence score includes a transformation and wherein said transformation is at least one of the following types of geometric transformations: identity, translation, rigid, rotation, reflection, scaling, affine, projective, and non-rigid.

15. The method according to claim 1 and wherein said global evidence score (GES) is defined as at least one of: GES ⁢ ( Q | H ref ) = log ⁢ ⁢ P ⁡ ( Q | H ref ) P ⁡ ( Q | H 0 ) = log ⁢ ⁢ ∑ S ⁢ P ⁡ ( Q | H ref , S ) ⁢ P ⁡ ( S | H ref ) P ⁡ ( Q | H 0 ) = log ⁢ ⁢ ∑ S ⁢ P ⁡ ( S | H ref ) ⁢ ⁢ P ⁡ ( Q | H ref , S ) P ⁡ ( Q | H 0 ) where P indicates a probability function, H ref is the hypothesis that a region R was generated by reference REF, H 0 is the null hypothesis, Q is said query signal and S is a segmentation of Q; and also comprising: computing at least one of an upper bound on GES(Q|H ref ) and a lower bound on GES(Q|H ref ).

16. The method according to claim 1 wherein said signals are digital media and also comprising: generating importance scores of at least portions of query digital media as a function of at least one of local evidence scores and global evidence scores with respect to said reference digital media; and enabling exploring said query digital media according to said importance scores.

17. The method according to claim 16 and wherein said importance scores are at least one of the following: user defined, pre-defined, defined by saliency within said query digital media, defined by similarity to objects, defined by similarity to actions and defined by similarity to speech and defined by irregularity in said query digital media.

18. The method according to claim 17 and comprising measuring said irregularity relative to one of the following: preceding data within said query digital media, present data within said query digital media, the entire query digital media, other portions of data within said query digital media and external reference digital media.

19. The method according to claim 16 and wherein said enabling comprises at least one of: enabling navigating through said query digital media as a function of at least one of importance scores, associations, labels and markings in said digital media, enabling moving through said query digital media at adaptive speeds as a function of the importance scores of said query digital media, enabling jumping to another interesting part of said query digital media according to said importance scores and enabling highlighting important regions in said query digital media, wherein said important regions have high importance scores.

20. The method according to claim 16 and wherein said enabling comprises enabling focusing on interesting parts of a current portion of said query digital media, wherein said interesting parts are defined by said importance scores.

21. The method according to claim 20 and wherein said enabling focusing comprises at least one of: cropping the uninteresting parts of said current portion, changing the temporal length of the video sequence, generating a shorter video clip, playing only said interesting parts of said current portion, omitting the uninteresting parts of said current portion and changing the display size to one of the following: the size of a mobile phone screen, a movie screen, a TV screen, an MP player screen, a portable screen, a thumbnail and an image preview.

22. The method according to claim 20 and also comprising generating one of the following: a video thumbnail, a video clip, a video summary, a movie trailer, a video synopsis, a video preview, a video index and a story board.

23. The method according to claim 16 and wherein said enabling comprises also ranking portions of said query digital media as a function of said importance scores.

24. The method according to claim 23 and also comprising sorting items as a function of said ranking, wherein said items are one of the following: image files, video sequence files, audio files, file directories, files found on the internet and files retrieved by a search engine.

25. The method according to claim 16 wherein said query digital media is a set of items, wherein said items are at least one of: images and video sequences, and also comprising extracting important items from said query digital media, wherein said important items have high importance scores.

26. The method according to claim 25 and also comprising generating a summary of a file directory, wherein said file directory is said query digital media and said summary is said important items.

27. The method according to claim 26 and also comprising generating from said important items at least one of the following: a video summary, a video preview, a video synopsis and a video directory summary.

28. The method according to claim 16 and wherein said query digital media is one portion of input digital media and said reference digital media is another portion of said input digital media.

29. The method according to claim 1 wherein said signals are digital media and also comprising providing associations from said reference signal to regions of said query signal which have high local evidence scores.

30. The method according to claim 29 and also comprising using said associations for at least one of the following: enabling navigating through said query digital media, propagating markers within said digital media and propagating meta-data within said digital media.

31. The method according to claim 29 and wherein said query digital media is one portion of input digital media and said reference digital media is another portion of said input digital media.

32. A method for exploring media, the method comprising: generating importance scores of at least portions of query digital media as a function of at least one of local evidence scores and global evidence scores, wherein said local evidence scores are functions at least of a quality of the match of regions of reference digital media to regions of said query digital media and of their statistical significance, and wherein said global evidence score defines the extent of a global similarity of a portion of said query digital media to said reference digital media; and enabling exploring said query digital media according to said importance scores, wherein said generating comprises: for at least one point q in said query digital media, determining one or more candidate regions in said reference digital media which are similar to a region in the vicinity of said point q; for each said candidate region computing a local evidence score of said region; and associating with said region in the vicinity of said point q a local evidence score which is a function of at least one of said local evidence scores of said candidate regions.

33. The method according to claim 32 and wherein said digital media comprises one of the following types of media: images, video sequences and audio data.

34. The method according to claim 32 and wherein said enabling comprises at least one of: navigating through said query digital media as a function of at least one of importance scores, associations, labels and markings in said digital media; playing through said query digital media at adaptive speeds as a function of the importance scores of said query digital media; jumping to another interesting part of said query digital media according to said importance scores; and highlighting important regions in said query digital media, wherein said important regions have high importance scores.

35. The method according to claim 32 and wherein said enabling also comprises focusing on interesting parts of a current portion of said query digital media, wherein said interesting parts are defined by said importance scores.

36. The device The method according to claim 35 and wherein said focusing comprises at least one of: cropping the uninteresting parts of said current portion; changing the temporal length of the video sequence; playing only said interesting parts of said current portion; and changing the display size to one of the following: the size of a mobile phone screen, a movie screen, a TV screen, an MP player screen, a portable screen, a thumbnail and an image preview.

37. The method according to claim 35 and also comprising generating one of the following: a video thumbnail, a video clip, a video summary, a movie trailer, a video synopsis, and a video preview.

38. The method according to claim 32 and wherein said enabling also comprises ranking portions of said query digital media as a function of said importance scores.

39. The method according to claim 38 and also comprising sorting items as a function of said ranking, wherein said items are one of the following: image files, video sequence files, audio files, file directories, files found on the internet and files retrieved by a search engine.

40. The method according to claim 32 wherein said query digital media is a set of items, wherein said items are at least one of: images and video sequences, and also comprising extracting important items from said query digital media, wherein said important items have high importance scores.

41. The method according to claim 40 and also comprising generating a summary of a file directory, wherein said file directory is said query digital media and said summary is said important items.

42. The method according to claim 32 and wherein said importance scores are at least one of the following: user defined, pre-defined, defined by saliency within said query digital media, defined by similarity to objects, defined by similarity to actions and defined by similarity to speech and defined by irregularity in said query digital media.

43. The method according to claim 32 and wherein said query digital media is one portion of input digital media and said reference digital media is another portion of said input digital media.

44. The method according to claim 32 and wherein said determining comprises: finding candidate small regions in said reference digital media which are similar to a small region in the vicinity of said point q; maximally expanding each said candidate region while increasing said local evidence score of said region; and choosing for said point q its region as the expanded region with the maximal said local evidence score.